Probabilistic ML & Uncertainty

Research / 2025 — 2026

Probabilistic ML & Uncertainty

An experimental extension studying Bayesian Linear Regression and Gaussian Processes under in-distribution and out-of-distribution conditions.

Select an image to view it full size.

My contribution

Extended the original project with reproducible experiments, explicit distribution splits, calibration measures, and visual diagnostics.

The approach

• Deterministic data generation and reproducible evaluation.
• Compare RMSE, negative log likelihood, predictive interval coverage, and calibration.
• Study how kernel choices change uncertainty under distribution shift.

Scope & perspective

Good calibration on familiar data does not guarantee reliable uncertainty when the distribution changes.

Original project by Mukul Kashyap; experimental enhancements by Arnav Goyal.